Chat Picker

ChatGPT vs Claude: Testing Image Understanding and Multimodal Tasks

Vendor pages document different upload rules and data controls, so this same-input plan compares ChatGPT and Claude without borrowed scores or claimed results.

Sources checked 2 Oct 2026

Vendor pages document what ChatGPT and Claude accept, but they do not supply a shared image-understanding result. Their upload guides set out different limits for OpenAI, Anthropic, and Google. Chat Picker has not tested the assistants, so use this same-input plan and the sourcing method rather than an external score.

Know these limits before you start

Check these published limits, as read on the vendor pages in October 2026:

  • ChatGPT: OpenAI says uploads are available on Free and paid plans, subject to plan limits and settings. Images are capped at 20 MB each, Free users have 3 file uploads per day, and Free limits may fall during peak hours. (OpenAI File Uploads FAQ, read October 1, 2026.)
  • Claude: Anthropic lists a 500 MB per-file chat-upload limit and a 30 MB per-project file limit. For PDFs of 100 pages or fewer, it analyzes text and visual elements; from 101 to 1000 pages, it processes text only. Non-PDF documents are text-only and cannot read embedded images. (Anthropic's upload guide, read October 1, 2026.)
  • Gemini: Google says a prompt can contain up to 10 supported files, subject to availability. Each video can be up to 2 GB, while other supported files can be up to 100 MB. (Google's file-upload guide, read October 1, 2026.)

Before uploading, review data settings. OpenAI says a training opt-out is available on Free, Go, Plus, and Pro. Anthropic says the opt-out applies to Free, Pro, and Max, while Team is not trained on by default. The Google AI plans page did not state a training setting when read on October 1, 2026, so check your current account controls.

Use material you are authorized to test, preferably synthetic or redacted samples. OpenAI's usage policies prohibit unauthorized aggregation or profiling of private or sensitive information. Anthropic's Usage Policy calls for relevant human expertise in elevated-risk uses and AI disclosure when outputs go directly to consumers. Google's Gemini guidelines address potentially harmful inaccuracies and say LLMs are probabilistic.

The test plan

Run each task in a fresh chat. Keep a log with the input type, plan, visible model choice, elapsed time, errors, limit messages, and price shown. Use identical files and instructions; leave model selection at its default when a selector appears.

  1. Documents, tables, and handwriting. Prepare a receipt with a clear total, a table with merged headers, and a handwritten note with a date and amounts. Upload each item separately and use this prompt: “Transcribe all visible text exactly, preserve row and column relationships, and identify uncertain handwriting without guessing.” Compare every field with your prepared source. Log wrong characters, totals, omitted rows, and whether uncertainty is stated.

  2. Charts and graphs. Prepare a bar or line chart with labeled axes, units, a legend, and a marked outlier. Ask: “Extract the plotted values, state the main pattern, and separate what the chart shows from what it only suggests.” Look for wrong values, swapped series, ignored units, and unsupported causal claims. Record each discrepancy.

  3. Screenshots, UI mockups, and cross-image evidence. Prepare a dense application screenshot showing a total, a UI mockup, and a receipt carrying the same total. Upload the screenshot and mockup, then ask: “Identify the visible controls, transcribe the form labels and error message, and explain which elements overlap or sit inside the modal.” Check small text, reading order, and spatial relationships. Add the receipt and ask: “Reconcile the receipt total with the application total and show the matching values.” Log invented controls, missed constraints, and arithmetic errors.

  4. Latency, cost, and throughput. Keep the device, connection, app or browser version, account plan, and default setting as constant as possible. Record elapsed time from submission to completion, retries, limit messages, and usage counters when shown. For throughput, submit the prepared files separately and note where access stops. Record the displayed subscription price, but do not derive a cost per answer.

  5. Failure modes and workflow fit. Save every failure. Start a fresh chat with the same image, paste the previous answer, and ask: “Identify every statement that conflicts with the image, quote the visible evidence for each correction, and list anything the image does not establish.” Log omitted details, misread text, arithmetic errors, spatial errors, cross-image errors, refusals, and source or tool issues. Record any new error introduced by the correction.

How to read your results

Set pass conditions as you prepare the files. Treat exact totals, required controls, and correct cross-image reconciliation as hard conditions; treat wording and chart style as preferences. Review hard failures first, then observed time and limit messages, then data controls and subscription price. If an assistant misses a hard condition, do not use it as the sole path for that task. If both pass, select based on your own workload; if neither does, retain a manual or human-reviewed step. Do not generalize a single workload's log into a universal ranking.

Where the plans differ

Use the Claude vs ChatGPT vs Gemini matrix to check the rows named Free plan, Low-cost tier, Main paid plan, Heavy-use plans, Team plan, Context window in the app, How usage limits are described, Training on your chats, and Ads. Each matrix figure carries its vendor source and read date.

  • Main paid plan: As read on October 1, 2026, ChatGPT Plus is $20/month billed monthly (OpenAI); Claude Pro is $20 monthly or $17/month on annual billing, with $200 up front (Anthropic); and Google AI Pro is $19.99/month (Google).
  • How usage limits are described: ChatGPT and Claude publish no exact message counts on their pricing pages. Gemini uses compute-based limits that refresh every 5 hours up to a weekly limit, with paid tiers listed at 2x, 4x, and up to 20x. These pricing pages were read October 1, 2026.
  • Context window in the app: ChatGPT lists separate instant and reasoning figures by plan. Claude says it offers up to 1M on every plan, varying by model, while Google's plans page gives no app token figure. (OpenAI pricing, Claude pricing, and Google AI plans, read October 1, 2026.)
  • Before subscribing: If the throughput test reaches a limit, check Heavy-use plans. For a work account, check Team plan. The Ads row notes that ChatGPT Go may include ads, while ad status for Claude and Gemini was not verified on the pricing pages read.

Sources

For current prices and limits, see the dated comparison pages.

Compare assistants